AMD Helios: 72× MI455X racks, Anthropic up to 2GW

What was reported

On 23 July 2026, Advanced Micro Devices (AMD) launched the AMD Helios rack-scale system at Advancing AI 2026. The company’s investor release describes Helios as 72 AMD Instinct MI455X graphics processing units (GPUs) and 18 sixth-generation AMD EPYC “Venice” central processing units (CPUs), linked with AMD Pensando networking and AMD ROCm software.

AMD said Helios is in production for gigawatt-scale deployment. That is a vendor tense. OpenAI, in the same 23 July 2026 release, said it expects to bring Helios online beginning in the fourth quarter of 2026, with deployments accelerating through 2027. Treat Q4 2026 as OpenAI’s stated start, not as a completed EU cluster.

AMD claims up to 30 percent more tokens per dollar than “the leading competitive solution.” The footnote in the AMD release, not repeated in full here, is the place to read the test setup. AGICY does not restate that 30 percent as an independent benchmark.

Google News still clustered Helios items through 12 August 2026, including an EE Times Asia headline that AMD launched Helios as a rack-scale AI infrastructure solution, and a StorageNewsletter item that Cirrascale said it is advancing open AI infrastructure with Helios and Instinct MI400-series GPUs. Those headlines are in the scrape. Full Cirrascale contract terms are not in the sources used for this briefing.

Anthropic: up to 2 GW, first gigawatt in H1 2027

On 22 July 2026, AMD and Anthropic said Anthropic will deploy up to 2 gigawatts (GW) of AMD Instinct MI450-series GPUs in Helios racks. The first gigawatt is due in the first half of 2027. The GPUs named for the racks are Instinct MI455X, part of the MI450 series, with EPYC Venice CPUs, Pensando networking, and ROCm. The deal builds on Anthropic’s existing use of Instinct MI355X GPUs.

AMD committed to a strategic equity investment of up to $5 billion in Anthropic. Datacenter Dynamics (DCD) reported the same figures and said AMD will begin shipping the rack-scale solution to customers, including Microsoft, in the second half of 2026. DCD also said Microsoft would use Helios for model inference internally and for Azure AI customers. The AMD and DCD texts do not name a European Union (EU) site for the 2 GW.

The companies said they will use Claude to optimize Instinct workloads and to accelerate ROCm development. AMD said it will adopt Claude across engineering and product teams. That is a software-collaboration clause, not a power-purchase agreement.

We are thrilled to deepen our partnership with Anthropic and deploy AMD Helios at gigawatt scale.

That sentence is from AMD chief executive Dr. Lisa Su, as quoted by DCD. Location, tariff, and grid queue are not in the quote.

Do not confuse AMD Helios with Helios Nordic Energy

On 14 August 2026, DCD also reported that Helios Nordic Energy Finland Oy, a VINCI-owned solar developer, filed a zoning initiative for an 80-hectare site at Alapitkä in Lapinlahti, Finland, next to a Fingrid substation, with reported potential of up to 500 MW. That Helios is not AMD Helios. The names collide in news clusters. They are different companies.

What it means for EU compute

AMD Helios is on the AGICY hardware watch as a rack-scale alternative to NVIDIA NVL-class systems. The July 2026 launch puts 72 MI455X GPUs in one advertised rack. The Anthropic 2 GW figure is a deployment ceiling with a first-gigawatt date in 2027. It is not live EU colocation.

A European buyer who wants Instinct hours in 2026 is still in the partner-rental window. GPU Bridge is hourly EU partner capacity. It is not a Helios hall in Vasilikos. The planned Cyprus campus is pre-construction. Do not treat AMD’s “in production” language as AGICY stock on the floor.

If Microsoft’s second-half 2026 shipping window holds, Helios becomes an Azure and hyperscale story before it becomes a Cyprus colocation story. The sources do not publish a euro-per-hour Helios list price.

GPU leasing, rent, buy, or colocation

AGICY paths for this silicon: GPU leasing, rent GPU server, buy AI accelerator (client-site finance), or AI colocation / GPU colo (also searched as collocation). Fleet in view: AMD Helios. Planned Cyprus campus — pre-construction, not a live hall. GPU leasing estimates · hardware roadmap · SRA.